A Lactylation-Related Radiosensitivity Index Predicts Differential Radiotherapy Benefit in Head and Neck Squamous Cell Carcinoma

A Lactylation-Related Radiosensitivity Index Predicts Differential Radiotherapy Benefit in Head and Neck Squamous Cell Carcinoma | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Lactylation-Related Radiosensitivity Index Predicts Differential Radiotherapy Benefit in Head and Neck Squamous Cell Carcinoma Wang Zhe, Xiong Yan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9250884/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Radiotherapy (RT) is integral to the management of head and neck squamous cell carcinoma (HNSCC), yet no validated biomarker exists to predict which patients benefit most. Lactylation, a metabolic post-translational modification, has been linked to DNA damage repair (DDR) and therapeutic resistance, but no lactylation-based radiosensitivity biomarker has been established in HNSCC. Here, we analyzed RNA-Seq and clinical data from 511 TCGA-HNSCC patients (308 RT, 203 non-RT) and constructed a three-gene lactylation-related radiosensitivity index (LRSI) comprising CARS2, KARS1, and BCL6 via treatment-stratified gene selection and LASSO-Cox regression. Radiosensitive (RS) patients derived significant RT benefit (HR = 0.45, 95% CI 0.30–0.67, P < 0.001), whereas radioresistant (RR) patients did not (HR = 0.98, P = 0.93; P_interaction = 0.0035). These results were robust across propensity score matching, inverse probability of treatment weighting, HPV-adjusted, and surgery/chemotherapy-adjusted analyses. In the independent GSE67614 post-RT cohort (n = 102), the LRSI discriminated locoregional recurrence with an AUC of 0.73 (adjusted OR = 2.05, P = 0.013). Biological characterization revealed that RR tumors exhibited elevated DDR activity, reduced immune infiltration, and heightened senescence-associated secretory phenotype. The LRSI is a predictive signature linking lactylation, DDR, and the tumor immune microenvironment to differential RT benefit in HNSCC. Prospective validation is warranted. Head and neck squamous cell carcinoma Lactylation Radiosensitivity Predictive biomarker Tumor immune microenvironment Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Head and neck squamous cell carcinoma (HNSCC) is the sixth most common malignancy worldwide, with approximately 890,000 new cases and 450,000 deaths reported annually [ 1 ]. Radiotherapy (RT), whether delivered as definitive treatment or after surgical resection, is integral to the management of HNSCC and improves both locoregional control and OS [ 2 , 3 ]. Yet clinical outcomes after RT remain highly variable: some patients develop locoregional recurrence despite adequate doses, while others endure treatment toxicities without commensurate benefit [ 4 , 5 ]. This heterogeneity underscores the need for biomarkers capable of identifying patients most likely to benefit from RT. Several gene expression-based radiosensitivity signatures have been proposed, the most widely studied being the radiosensitivity index (RSI) derived from pan-cancer cell line data [ 6 , 7 ]. However, most existing signatures were developed in uniformly irradiated populations without a non-treated control arm, rendering them prognostic rather than truly predictive [ 6 – 8 ]. Although Cui et al. demonstrated treatment-stratified predictive evaluation for a combined radiosensitivity–immune signature in breast cancer, such approaches remain rare in HNSCC[ 9 ]. A predictive biomarker must demonstrate a differential treatment effect—a statistically significant interaction between the biomarker and treatment assignment—rather than simply stratifying outcomes irrespective of therapy [ 8 ]. Few radiosensitivity signatures in HNSCC have been evaluated with interaction testing, propensity score adjustment, or causal inference methods. Lactylation has recently emerged as a post-translational modification linking cellular metabolism to epigenetic regulation [ 10 ]. Lactate, the end product of the Warburg effect, serves as a substrate for histone lysine lactylation (Kla), which reprograms gene expression programs involved in DDR, immune evasion, and senescence [ 11 , 12 ]. Notably, lactylation of the DDR protein NBS1 enhances homologous recombination repair and confers resistance to genotoxic therapies [ 13 ]. These observations suggest that lactylation-related gene expression may capture a metabolic–epigenetic axis relevant to radiation response, yet no such signature has been developed for HNSCC. Here we report the construction and evaluation of a lactylation-related radiosensitivity index (LRSI) designed to predict differential RT benefit in HNSCC. A treatment-stratified gene selection strategy was used to ensure predictive rather than merely prognostic specificity, and the index was evaluated using interaction testing, propensity score methods, RCS modeling, and an external post-RT cohort. We further characterized the biological basis of the LRSI through analyses of the lactylation–DDR axis, tumor immune microenvironment, SASP, and predicted drug sensitivity. Methods Data acquisition and preprocessing RNA-Seq raw count data and clinical information for 520 HNSCC patients were obtained from The Cancer Genome Atlas (TCGA) via the GDC Data Portal. Gene expression was normalized using the trimmed mean of M-values (TMM) method in edgeR and expressed as log 2 counts per million (logCPM; prior count = 1). Where multiple Ensembl IDs mapped to the same gene symbol, the mean expression value was retained. Survival endpoints were defined according to the TCGA Clinical Data Resource; treatment annotations including radiotherapy receipt were obtained from the TCGA clinical data as provided through the GDC Data Portal. After excluding patients with OS ≤ 30 days (n = 8) or missing survival data (n = 1), 511 patients remained for analysis (RT, n = 308; non-RT, n = 203). For external evaluation, the GSE67614 dataset (n = 102 high-risk HNSCC patients treated with postoperative RT) was downloaded from the Gene Expression Omnibus (GEO). Construction of the LRSI We assembled 547 candidate LRGs from eight Molecular Signatures Database (MSigDB) gene sets related to glycolysis, lactate metabolism, and lactic acidosis (HALLMARK_GLYCOLYSIS; GOMF_LACTATE_TRANSMEMBRANE_TRANSPORTER_ACTIVITY; HP_LACTIC_ACIDOSIS; HP_INCREASED_CIRCULATING_LACTATE_CONCENTRATION; HP_INCREASED_CIRCULATING_LACTATE_DEHYDROGENASE_CONCENTRATION; HP_ABNORMAL_CIRCULATING_LACTATE_DEHYDROGENASE_CONCENTRATION; KEGG_MEDICUS_REFERENCE_GLYCOLYSIS; WP_AEROBIC_GLYCOLYSIS), supplemented by a manually curated list of established lactylation writers, erasers, and readers [ 10 , 11 , 14 ]. Univariable Cox regression for OS was performed separately in the RT and non-RT cohorts; genes with false discovery rate (FDR) 0.20 in the non-RT cohort were retained as RT-specific candidates (n = 11). These were further filtered by DDR pathway correlation (|Spearman ρ| > 0.30), yielding four genes. Gene expression values were z-score standardized within the RT cohort before model fitting. LASSO-penalized Cox regression with ten-fold cross-validation (CV) was performed using the “glmnet” R package; λ 1se was selected to yield a parsimonious model. The resulting three-gene signature was: LRSI = 0.241 × CARS2 + 0.239 × KARS1 − 0.175 × BCL6 where expression values denote z-scored logCPM. Patients were classified as RS or RR using the RT-cohort median LRSI (− 0.021) as the cutoff. The same coefficients and cutoff were applied to the non-RT cohort. Predictive value assessment and robustness analyses KM analysis with log-rank tests compared OS between RS and RR within each treatment arm, and between RT and non-RT within each LRSI stratum. Cox models with a multiplicative RT × LRSI interaction term tested for treatment-effect modification. To address confounding by treatment indication, PSM and IPTW-ATT were performed within each LRSI stratum using the “MatchIt” and “WeightIt” R packages, adjusting for age, sex, stage, grade, smoking history, and human papillomavirus (HPV) status; covariate balance was verified by standardized mean differences (SMD). RCS Cox models evaluated the continuous relationship between LRSI and the RT effect. Further sensitivity analyses included multivariable Cox regression adjusting for HPV status alone, for surgery type, and for both surgery type and chemotherapy receipt. External cohort analysis In the GSE67614 cohort, the TCGA-derived LRSI coefficients were retained, while gene expression values were z-score normalized within the cohort to account for platform differences (microarray vs. RNA-Seq). Patients were dichotomized at the cohort-specific LRSI median. LRSI was assessed as both a continuous (per 1-SD) and dichotomized (RS/RR) predictor of locoregional recurrence. Discriminative ability was quantified by receiver operating characteristic (ROC) analysis using the “pROC” R package. Logistic regression models, with and without adjustment for T stage, N stage, and margin status, estimated the association between LRSI and recurrence risk. LRSI distributions were compared across failure phenotypes (no evidence of disease [NED], locoregional recurrence, and distant metastasis) using Kruskal-Wallis tests. Immune infiltration, SASP, and drug sensitivity analyses Immune cell abundance in the RT cohort was estimated using MCPcounter. A lactylation-informed DDR bridge score integrating glycolysis, hypoxia, and DNA repair activities was computed by single-sample gene set enrichment analysis (ssGSEA) using the “GSVA” R package. SASP activity was quantified via ssGSEA using senescence- and SASP-related gene sets (SenMayo, Reactome SASP, and Reactome cellular senescence). Predicted drug sensitivity was estimated with the “oncoPredict” R package using the Genomics of Drug Sensitivity in Cancer 2 (GDSC2) training panel; differential sensitivity between RS and RR was assessed by Wilcoxon tests with FDR correction. Statistical analysis Continuous variables were compared using Wilcoxon rank-sum or Kruskal-Wallis tests; categorical variables using chi-square or Fisher exact tests. Survival was analyzed with KM curves, log-rank tests, and Cox proportional hazards regression. All P values were two-sided (α = 0.05). Analyses were performed in R version 4.5.1. Results Identification of RT-specific lactylation-related genes and construction of the LRSI Of 520 TCGA-HNSCC patients, 511 met the inclusion criteria (RT, n = 308; non-RT, n = 203; 215 OS events; Fig. 1 A). Baseline characteristics are summarized in Table 1 . From 547 candidate LRGs, 11 were identified as RT-specific prognostic genes (FDR 0.20 in the non-RT cohort), of which four passed DDR pathway correlation filtering (|Spearman ρ| > 0.30; Fig. 1 B). LASSO-penalized Cox regression with λ 1se yielded a three-gene signature (Fig. 1 C–D): LRSI = 0.241 × CARS2 + 0.239 × KARS1 − 0.175 × BCL6 CARS2 and KARS1 carried positive coefficients (higher expression associated with radioresistance), while BCL6 carried a negative coefficient (higher expression associated with radiosensitivity). Patients were dichotomized at the RT-cohort median LRSI (− 0.021) into RS (n = 154) and RR (n = 154) groups (Fig. 1 E). Table 1 Baseline clinicopathological characteristics of the TCGA-HNSC derivation cohort according to radiotherapy status Characteristic Overall (N = 511) No RT (N = 203) RT (N = 308) P value SMD N 511 203 308 Age at diagnosis, mean (SD) 61.26 (11.80) 63.56 (12.51) 59.74 (11.06) < 0.001 0.323 Male sex, n (%) 378 (74.0) 135 (66.5) 243 (78.9) 0.003 0.281 Histologic grade, n (%) 0.001 0.443 G1 61 (11.9) 36 (17.7) 25 (8.1) G2 297 (58.1) 120 (59.1) 177 (57.5) G3 124 (24.3) 44 (21.7) 80 (26.0) G4 7 (1.4) 0 (0.0) 7 (2.3) GX 18 (3.5) 3 (1.5) 15 (4.9) Unknown 4 (0.8) 0 (0.0) 4 (1.3) AJCC stage, n (%) < 0.001 0.763 I 27 (5.3) 17 (8.4) 10 (3.2) II 70 (13.7) 51 (25.1) 19 (6.2) III 81 (15.9) 41 (20.2) 40 (13.0) IV 259 (50.7) 81 (39.9) 178 (57.8) Unknown 74 (14.5) 13 (6.4) 61 (19.8) Pathologic T stage, n (%) < 0.001 0.506 T1-2 180 (35.2) 95 (46.8) 85 (27.6) T3-4 269 (52.6) 98 (48.3) 171 (55.5) Unknown 62 (12.1) 10 (4.9) 52 (16.9) Pathologic N stage, n (%) < 0.001 0.384 N0-1 240 (47.0) 118 (58.1) 122 (39.6) N2-3 172 (33.7) 51 (25.1) 121 (39.3) Unknown 99 (19.4) 34 (16.7) 65 (21.1) Smoking history, n (%) 0.039 0.216 Never 114 (22.3) 43 (21.2) 71 (23.1) Ever 385 (75.3) 151 (74.4) 234 (76.0) Unknown 12 (2.3) 9 (4.4) 3 (1.0) HPV status, n (%) 0.051 0.227 Negative 409 (80.0) 170 (83.7) 239 (77.6) Positive 71 (13.9) 19 (9.4) 52 (16.9) Unknown 31 (6.1) 14 (6.9) 17 (5.5) Values are presented as mean (SD) or n (%). P values and SMDs were recalculated for the compact main-text version. T stage was collapsed as T1-2, T3-4, and Unknown; N stage as N0-1, N2-3, and Unknown; smoking history as Never, Ever, and Unknown. LRSI predicts differential RT benefit Within the RT cohort, RR patients had significantly worse OS than RS patients (Fig. 2A; HR = 2.04, RR vs. RS, 95% CI 1.40–2.97, P < 0.001). No such difference was observed in the non-RT cohort (Fig. 2B; HR = 0.85, 95% CI 0.56–1.27, P = 0.41), indicating that the prognostic value of the LRSI was confined to RT-treated patients. When stratified by LRSI group, RS patients derived substantial OS benefit from RT (Fig. 2C; HR = 0.45, 95% CI 0.30–0.67, P < 0.001), whereas RR patients did not (Fig. 2D; HR = 0.98, 95% CI 0.68–1.42, P = 0.93). A significant multiplicative interaction between RT and LRSI group confirmed the predictive nature of this association ( P interaction = 0.0035). Robustness of the predictive effect After IPTW-ATT weighting in the RS subgroup, all covariate SMDs fell below 0.10 (Fig. 3A) and the RT benefit remained significant (Fig. 3B; P = 0.002). The RS subgroup showed consistent RT benefit across all analytical approaches: unadjusted HR = 0.45 (95% CI 0.30–0.67); propensity-based methods incorporating age, sex, stage, grade, smoking, and HPV status yielded PSM HR = 0.49 (95% CI 0.32–0.77) and IPTW-ATT HR = 0.46 (95% CI 0.28–0.74); multivariable Cox regression adjusting for HPV status gave HR = 0.40 (95% CI 0.26–0.62); adjustment for surgery type yielded HR = 0.35 (95% CI 0.21–0.58); and a fully adjusted model including both surgery and chemotherapy receipt yielded HR = 0.30 (95% CI 0.18–0.52). In the RR subgroup, no significant RT benefit was detected under any approach (unadjusted HR = 0.98, IPTW-ATT HR = 0.96, PSM HR = 0.84; all P > 0.20; Fig. 3C). RCS analysis confirmed a continuous decrease in RT benefit with increasing LRSI, with HR < 1 confined to the lower end of the distribution (spline interaction LRT P = 0.016; Fig. 3D). Association of LRSI with locoregional failure in an independent post-RT cohort To evaluate generalizability, we applied the TCGA-derived LRSI coefficients to the GSE67614 cohort (n = 102 high-risk HNSCC patients treated with postoperative RT). The LRSI discriminated locoregional recurrence with an AUC of 0.73 (95% CI 0.62–0.84; Fig. 4 A). Patients who developed locoregional recurrence had significantly higher LRSI scores (Fig. 4 B; P = 0.001). Each 1-SD increase in LRSI was associated with a 2.10-fold higher recurrence risk (OR = 2.10, 95% CI 1.23–3.58, P = 0.006), and this association persisted after adjustment for T stage, N stage, and margin status (adjusted OR = 2.05, 95% CI 1.16–3.61, P = 0.013; Fig. 4 C). Dichotomized LRSI showed a 4.21-fold increased recurrence risk in RR versus RS patients (OR = 4.21, 95% CI 1.41–12.59, P = 0.010). The association was most pronounced for locoregional recurrence across failure phenotypes (Kruskal-Wallis P = 0.006; Fig. 4 D), although the small number of distant metastasis events limits conclusions regarding endpoint specificity. Biological characterization of LRSI-defined subgroups RR tumors exhibited significantly higher DDR bridge scores, reflecting elevated glycolysis–hypoxia–DNA repair activity (Fig. 5 A; P < 0.001). MCPcounter-based deconvolution revealed that RS tumors harbored greater infiltration of myeloid dendritic cells (DCs), B-lineage cells, T cells, and endothelial cells (all P < 0.01; Fig. 5 B). SASP activity was elevated in RR tumors in both the RT and non-RT cohorts (both P < 0.001; Fig. 5 C). Drug sensitivity prediction identified subgroup-specific vulnerabilities: RS tumors showed greater predicted sensitivity to dasatinib, luminespib, and trametinib, whereas RR tumors showed greater sensitivity to docetaxel, obatoclax mesylate, and ulixertinib (FDR-corrected; Fig. 5 D). Discussion The LRSI reported here is a three-gene expression signature that identifies HNSCC patients who derive differential benefit from RT. Constructed through RT-specific gene selection, DDR pathway coupling, and LASSO regularization, the model comprising CARS2 , KARS1 , and BCL6 distinguished an RS subgroup with significant RT benefit from an RR subgroup without benefit. The significant treatment–biomarker interaction and robustness across multiple confounding-adjustment methods—including PSM, IPTW, HPV-adjusted, and surgery/chemotherapy-adjusted models—support the LRSI as a candidate predictive biomarker in this disease. Separately, elevated LRSI was associated with locoregional failure in an independent post-RT cohort, providing prognostic corroboration in a distinct clinical setting. Distinguishing predictive from prognostic biomarkers remains a central challenge [ 8 , 15 ]. Many radiosensitivity gene signatures were developed in patients who all received RT, without a non-RT control group, and are therefore prognostic by design [ 6 – 8 , 16 ]. The LRSI was built to address this: genes were retained only if they carried prognostic significance in the RT cohort (FDR 0.20), thereby excluding genes with treatment-independent effects. The null association in non-RT patients and the significant interaction term reinforce the claim of predictive, rather than merely prognostic, utility. Existing radiosensitivity models for HNSCC, including the RSI and multi-gene signatures from radiation-response gene sets, were generally trained on pan-cancer cell line data or lacked treatment-stratified evaluation, limiting their use as predictive tools[ 7 , 17 , 18 ]. The LRSI differs in being derived from a biologically defined candidate set—lactylation-related genes—rather than an agnostic transcriptome-wide screen, and in being evaluated through causal inference methods (PSM, IPTW) and continuous dose–response modeling (RCS). The spline analysis demonstrated a graded decrease in RT benefit with rising LRSI, arguing against a purely artifactual dichotomous effect and strengthening the case for clinical relevance. CARS2 encodes mitochondrial cysteinyl-tRNA synthetase, which produces cysteine persulfides essential for electron transport chain function [ 19 ]. Aminoacyl-tRNA synthetases have recently been recognized as non-canonical regulators of lactylation—AARS1, for instance, acts as a lactate sensor and lactyltransferase—and the positive LRSI coefficient of CARS2 is consistent with the idea that enhanced mitochondrial metabolic capacity contributes to radioresistance[ 20 ]. KARS1 (lysyl-tRNA synthetase) has been linked to colorectal cancer diagnosis and promotes metastasis through the KARS1/ERK/paxillin axis while polarizing macrophages toward an M2 phenotype, an immunosuppressive property that aligns with the lower immune infiltration we observed in RR tumors[ 21 , 22 ]. BCL6 is a transcriptional repressor of ATR and p53 that enables genotoxic stress evasion in solid tumors [ 23 – 25 ]. Its negative coefficient—meaning higher expression is associated with radiosensitivity—may reflect DDR attenuation. Whether this link additionally involves B-cell-mediated immunity remains to be resolved at single-cell resolution. The lactylation–DDR connection offers a biologically coherent framework for the LRSI. Histone Kla, driven by aerobic glycolysis, reprograms gene expression toward DNA repair and immune evasion, and lactylation of NBS1 has been shown to directly promote homologous recombination and chemoresistance [ 10 , 11 , 13 ]. Consistent with this, RR tumors in our cohort had significantly higher DDR bridge scores integrating glycolysis, hypoxia, and DNA repair pathway activity, supporting a model in which a lactylation-fueled metabolic-repair program underlies the radioresistant phenotype. The tumor immune microenvironment increasingly appears to modulate radiation response [ 5 ]. RS tumors harbored greater infiltration of myeloid DCs, B-lineage cells, T cells, and endothelial cells, suggesting an immunologically active milieu that may potentiate the in situ vaccination effects of RT [ 26 ]. Conversely, RR tumors showed elevated SASP activity—a phenotype linked to NF-κB-dependent cytokine-mediated radioresistance in HNSCC [ 27 ]. The convergence of immune exclusion and SASP elevation in RR tumors points to a pro-tumorigenic microenvironment that blunts radiation efficacy. Drug sensitivity prediction revealed subgroup-specific therapeutic vulnerabilities: RS tumors showed greater predicted sensitivity to dasatinib, luminespib (HSP90 inhibitor), and trametinib (MEK inhibitor), whereas RR tumors showed greater sensitivity to docetaxel and ulixertinib (ERK inhibitor). Although hypothesis-generating, these patterns raise the possibility that the LRSI could inform not only RT selection but also the choice of concurrent or adjuvant systemic therapy. In the GSE67614 cohort, applying the TCGA-derived LRSI coefficients with cohort-internal z-score standardization yielded acceptable discrimination for locoregional recurrence and maintained significance after adjustment for clinicopathological factors. The association was strongest for locoregional failure, consistent with the expectation that the signature captures local radiation response, though the small number of distant metastasis events limits conclusions on endpoint specificity. Because GSE67614 comprises uniformly RT-treated patients, it cannot directly test treatment–biomarker interaction; rather, these data demonstrate that elevated LRSI is associated with locoregional failure in an independent post-RT population. Several limitations should be acknowledged. First, this study was based on retrospective cohorts with non-randomized treatment assignment; despite PSM and IPTW, residual unmeasured confounding cannot be excluded, and predictive biomarkers ideally require validation in prospective randomized trials [ 8 , 15 ]. Second, the LRSI was derived from bulk RNA-Seq, which does not capture intratumoral heterogeneity or spatial context; single-cell and spatial transcriptomic studies may offer finer resolution. Third, the external cohort used a different endpoint (locoregional recurrence) and platform (microarray), and further validation in independent cohorts with time-to-event endpoints is needed. Fourth, our interpretation of the lactylation–DDR axis relies on pathway-level inference without direct measurement of protein lactylation; experimental validation is required. Finally, the TCGA cohort is predominantly of European descent, and generalizability to other populations warrants investigation. Conclusions The LRSI ( CARS2 , KARS1 , BCL6 ) is a three-gene signature that identifies HNSCC patients with differential RT benefit (RS: HR = 0.45; RR: HR = 0.98; P interaction = 0.0035). This predictive effect was robust across multiple confounding-adjustment approaches. In an independent post-RT cohort, elevated LRSI was associated with locoregional failure (AUC = 0.73), though this uniformly treated cohort could not directly assess treatment–biomarker interaction. The signature captures a convergent axis linking glycolysis–lactylation–DDR activity, immune microenvironment composition, and SASP to radiation response. Prospective validation in randomized trials is warranted to establish clinical utility and guide personalized RT strategies in HNSCC. Declarations Ethics approval and consent to participate All data were obtained from publicly available databases (TCGA and GEO). No additional ethical approval was required. Availability of data and materials TCGA-HNSCC data are available from the GDC Data Portal (https://portal.gdc.cancer.gov/). The GSE67614 dataset is available from GEO (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE67614). Competing interests The authors declare no competing interests. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Authors’ contributions Xiong Yan: Conceptualization, Methodology, Formal analysis, Writing – original draft, Writing – review & editing, Supervision. Wang Zhe: Data curation, Software, Validation, Visualization, Writing – review & editing. Equal contribution: Xiong Yan and Wang Zhe contributed equally to this work and share first authorship. Acknowledgements We acknowledge TCGA, and GEO for providing access to the datasets used in this study. References Sung H, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209–49. Bernier J, et al. Postoperative irradiation with or without concomitant chemotherapy for locally advanced head and neck cancer. N Engl J Med. 2004;350(19):1945–52. Cooper JS, et al. Postoperative concurrent radiotherapy and chemotherapy for high-risk squamous-cell carcinoma of the head and neck. N Engl J Med. 2004;350(19):1937–44. Johnson DE, et al. Head and neck squamous cell carcinoma. Nat Rev Dis Primers. 2020;6(1):92. McLaughlin M, et al. Inflammatory microenvironment remodelling by tumour cells after radiotherapy. Nat Rev Cancer. 2020;20(4):203–17. Torres-Roca JF, et al. Prediction of radiation sensitivity using a gene expression classifier. Cancer Res. 2005;65(16):7169–76. Eschrich SA, et al. A gene expression model of intrinsic tumor radiosensitivity: prediction of response and prognosis after chemoradiation. Int J Radiat Oncol Biol Phys. 2009;75(2):489–96. Ballman KV. Biomarker: Predictive or Prognostic? J Clin Oncol. 2015;33(33):3968–71. Cui Y, et al. Integrating Radiosensitivity and Immune Gene Signatures for Predicting Benefit of Radiotherapy in Breast Cancer. Clin Cancer Res. 2018;24(19):4754–62. Zhang D, et al. Metabolic regulation of gene expression by histone lactylation. Nature. 2019;574(7779):575–80. Chen J, et al. Lactate and lactylation in cancer. Signal Transduct Target Ther. 2025;10(1):38. Zhu C, et al. Lactylation in radiosensitivity regulation: Mechanistic insights and recent advances. Semin Oncol. 2025;52(5):152394. Chen H, et al. NBS1 lactylation is required for efficient DNA repair and chemotherapy resistance. Nature. 2024;631(8021):663–9. Wang J, et al. Ubiquitous protein lactylation in health and diseases. Cell Mol Biol Lett. 2024;29(1):23. Simon RM, Paik S, Hayes DF. Use of archived specimens in evaluation of prognostic and predictive biomarkers. J Natl Cancer Inst. 2009;101(21):1446–52. Eschrich SA, et al. Validation of a radiosensitivity molecular signature in breast cancer. Clin Cancer Res. 2012;18(18):5134–43. Lu C, et al. Construction and validation of a prognostic nine-gene signature associated with radiosensitivity in head and neck squamous cell carcinoma. Clin Transl Radiat Oncol. 2023;43:100686. You P, et al. Radiation-sensitive genetic prognostic model identifies individuals at risk for radiation resistance in head and neck squamous cell carcinoma. J Cancer Res Clin Oncol. 2023;149(17):15623–40. Akaike T, et al. Cysteinyl-tRNA synthetase governs cysteine polysulfidation and mitochondrial bioenergetics. Nat Commun. 2017;8(1):1177. Zong Z, et al. Alanyl-tRNA synthetase, AARS1, is a lactate sensor and lactyltransferase that lactylates p53 and contributes to tumorigenesis. Cell. 2024;187(10):2375–e239233. Suh JH et al. Plasma Lysyl-tRNA Synthetase 1 (KARS1) as a Novel Diagnostic and Monitoring Biomarker for Colorectal Cancer. J Clin Med, 2020. 9(2). Nam SH, et al. Lysyl-tRNA synthetase-expressing colon spheroids induce M2 macrophage polarization to promote metastasis. J Clin Invest. 2018;128(11):5034–55. Ranuncolo SM, et al. Bcl-6 mediates the germinal center B cell phenotype and lymphomagenesis through transcriptional repression of the DNA-damage sensor ATR. Nat Immunol. 2007;8(7):705–14. Phan RT, Dalla-Favera R. The BCL6 proto-oncogene suppresses p53 expression in germinal-centre B cells. Nature. 2004;432(7017):635–9. Liu Y et al. The oncoprotein BCL6 enables solid tumor cells to evade genotoxic stress. Elife, 2022. 11. Ngwa W, et al. Using immunotherapy to boost the abscopal effect. Nat Rev Cancer. 2018;18(5):313–22. Schoetz U, et al. Early senescence and production of senescence-associated cytokines are major determinants of radioresistance in head-and-neck squamous cell carcinoma. Cell Death Dis. 2021;12(12):1162. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 13 Apr, 2026 Editor invited by journal 03 Apr, 2026 Editor assigned by journal 28 Mar, 2026 Submission checks completed at journal 28 Mar, 2026 First submitted to journal 28 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9250884","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":622132121,"identity":"46337785-3af8-4a16-af99-bae977c2d62b","order_by":0,"name":"Wang Zhe","email":"","orcid":"","institution":"Sun Yat-sen University Cancer Center","correspondingAuthor":false,"prefix":"","firstName":"Wang","middleName":"","lastName":"Zhe","suffix":""},{"id":622132122,"identity":"015b1081-2f08-4f97-b9ed-3c388fa759e0","order_by":1,"name":"Xiong Yan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6UlEQVRIiWNgGAWjYDACZijNT7oWyQaSbTM4QLTK48zHHn5ts8szPn/G+DNPjR0Dv/TxCww/d+DWItnMlm4s25ZcbHYjx0ya51gyg2RfTgFj7xncWviZecykJduYE7fd4DFj5m04wGBwhieBmbENtxY2iJb6xM39QIcRpQVki+THtsOJGxhyDKQhWtgP4NUC9EuaNMO544kzbqSVSc45lswj2cPDcLAXjxaD84ePSf4oq07s7z+8+cObGjs5fh72hw9+4tECAsy8bCCKwwBE8gAR4Thi/PEHRLE/gPLhjFEwCkbBKBgFYAAADR1JjKC6waIAAAAASUVORK5CYII=","orcid":"","institution":"Sun Yat-sen University Cancer Center","correspondingAuthor":true,"prefix":"","firstName":"Xiong","middleName":"","lastName":"Yan","suffix":""}],"badges":[],"createdAt":"2026-03-28 08:08:49","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9250884/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9250884/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107377753,"identity":"5c31770e-5f9f-4d85-92f4-8c9d7b17ab7f","added_by":"auto","created_at":"2026-04-21 01:31:19","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":418330,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy design, gene selection funnel, and LRSI construction.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Patient selection flowchart from TCGA-HNSCC. (B) Stepwise gene filtering: 547 LRG candidates → 11 RT-specific genes → 4 DDR-coupled genes → 3 LRSI genes. (C) LASSO CV curve with λ\u003csub\u003emin\u003c/sub\u003e and λ\u003csub\u003e1se\u003c/sub\u003e indicated. (D) LASSO-Cox coefficients for the three signature genes. (E) LRSI score distribution in the RT and non-RT cohorts with the median cutoff (−0.021) indicated.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9250884/v1/01047df674f55d85deb08c2c.jpeg"},{"id":107485713,"identity":"b006faf6-17e3-4865-9c52-2eb6bd0dfcee","added_by":"auto","created_at":"2026-04-22 02:35:50","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":62752,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eLRSI predicts differential RT benefit.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eKM curves for OS: (A) RS vs. RR in the RT cohort; (B) RS vs. RR in the non-RT cohort; (C) RT vs. non-RT in the RS subgroup; (D) RT vs. non-RT in the RR subgroup. \u003cem\u003eP\u003c/em\u003e values from log-rank tests.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9250884/v1/7fe03bb9505b40368531e6b8.png"},{"id":107377755,"identity":"c5511f4f-65ca-40ef-be41-6bea008f2f44","added_by":"auto","created_at":"2026-04-21 01:31:19","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":93312,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRobustness analyses.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Love plot of covariate SMDs before and after IPTW weighting in the RS subgroup. (B) IPTW-weighted KM curves for RT vs. non-RT in the RS subgroup. (C) Forest plot summarizing RT benefit (HR with 95% CI) across analytical approaches in the RS and RR subgroups. (D) RCS curve depicting the continuous relationship between standardized LRSI and the HR for RT vs. non-RT.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9250884/v1/abe7eb2574bb86aebe692c22.png"},{"id":107377756,"identity":"5486fb16-8e80-47d6-bdbe-9b12e7abd45c","added_by":"auto","created_at":"2026-04-21 01:31:19","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":77076,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAssociation of LRSI with locoregional failure in the GSE67614 cohort. \u003c/strong\u003e(A) ROC curve for locoregional recurrence prediction (AUC = 0.73). (B) Box plot of LRSI scores by recurrence status. (C) Forest plot of logistic regression ORs (per 1-SD, RS vs. RR, and adjusted models). (D) LRSI distribution across failure phenotypes (NED, locoregional recurrence, distant metastasis).\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9250884/v1/e4d0b97d30aec60914e7f6af.png"},{"id":107488538,"identity":"df2e1885-bd5f-44d1-902f-e99b423e2497","added_by":"auto","created_at":"2026-04-22 02:45:01","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":393057,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBiological characterization of LRSI-defined subgroups.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Integrated DDR bridge score (glycolysis–hypoxia–DNA repair) in RS vs. RR within the RT cohort. (B) MCPcounter-estimated immune cell infiltration (myeloid DCs, B-lineage, T cells, endothelial cells) in RS vs. RR. (C) SASP activity (ssGSEA) in RS vs. RR within both RT and non-RT cohorts. (D) Volcano plot of predicted drug sensitivity differences (oncoPredict/GDSC2) between RS and RR in the RT cohort.\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9250884/v1/d6235bb1767943016eb20069.jpeg"},{"id":107489635,"identity":"539f787f-6558-4de0-bd09-9b2c115577bb","added_by":"auto","created_at":"2026-04-22 02:48:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1445472,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9250884/v1/2a71d7f8-dacc-42ac-b90f-6ac4d2991e3b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Lactylation-Related Radiosensitivity Index Predicts Differential Radiotherapy Benefit in Head and Neck Squamous Cell Carcinoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHead and neck squamous cell carcinoma (HNSCC) is the sixth most common malignancy worldwide, with approximately 890,000 new cases and 450,000 deaths reported annually [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Radiotherapy (RT), whether delivered as definitive treatment or after surgical resection, is integral to the management of HNSCC and improves both locoregional control and OS [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Yet clinical outcomes after RT remain highly variable: some patients develop locoregional recurrence despite adequate doses, while others endure treatment toxicities without commensurate benefit [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. This heterogeneity underscores the need for biomarkers capable of identifying patients most likely to benefit from RT.\u003c/p\u003e \u003cp\u003eSeveral gene expression-based radiosensitivity signatures have been proposed, the most widely studied being the radiosensitivity index (RSI) derived from pan-cancer cell line data [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. However, most existing signatures were developed in uniformly irradiated populations without a non-treated control arm, rendering them prognostic rather than truly predictive [\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Although Cui et al. demonstrated treatment-stratified predictive evaluation for a combined radiosensitivity\u0026ndash;immune signature in breast cancer, such approaches remain rare in HNSCC[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. A predictive biomarker must demonstrate a differential treatment effect\u0026mdash;a statistically significant interaction between the biomarker and treatment assignment\u0026mdash;rather than simply stratifying outcomes irrespective of therapy [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Few radiosensitivity signatures in HNSCC have been evaluated with interaction testing, propensity score adjustment, or causal inference methods.\u003c/p\u003e \u003cp\u003eLactylation has recently emerged as a post-translational modification linking cellular metabolism to epigenetic regulation [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Lactate, the end product of the Warburg effect, serves as a substrate for histone lysine lactylation (Kla), which reprograms gene expression programs involved in DDR, immune evasion, and senescence [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Notably, lactylation of the DDR protein NBS1 enhances homologous recombination repair and confers resistance to genotoxic therapies [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. These observations suggest that lactylation-related gene expression may capture a metabolic\u0026ndash;epigenetic axis relevant to radiation response, yet no such signature has been developed for HNSCC.\u003c/p\u003e \u003cp\u003eHere we report the construction and evaluation of a lactylation-related radiosensitivity index (LRSI) designed to predict differential RT benefit in HNSCC. A treatment-stratified gene selection strategy was used to ensure predictive rather than merely prognostic specificity, and the index was evaluated using interaction testing, propensity score methods, RCS modeling, and an external post-RT cohort. We further characterized the biological basis of the LRSI through analyses of the lactylation\u0026ndash;DDR axis, tumor immune microenvironment, SASP, and predicted drug sensitivity.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData acquisition and preprocessing\u003c/h2\u003e \u003cp\u003eRNA-Seq raw count data and clinical information for 520 HNSCC patients were obtained from The Cancer Genome Atlas (TCGA) via the GDC Data Portal. Gene expression was normalized using the trimmed mean of M-values (TMM) method in edgeR and expressed as log\u003csub\u003e2\u003c/sub\u003e counts per million (logCPM; prior count\u0026thinsp;=\u0026thinsp;1). Where multiple Ensembl IDs mapped to the same gene symbol, the mean expression value was retained. Survival endpoints were defined according to the TCGA Clinical Data Resource; treatment annotations including radiotherapy receipt were obtained from the TCGA clinical data as provided through the GDC Data Portal. After excluding patients with OS\u0026thinsp;\u0026le;\u0026thinsp;30 days (n\u0026thinsp;=\u0026thinsp;8) or missing survival data (n\u0026thinsp;=\u0026thinsp;1), 511 patients remained for analysis (RT, n\u0026thinsp;=\u0026thinsp;308; non-RT, n\u0026thinsp;=\u0026thinsp;203). For external evaluation, the GSE67614 dataset (n\u0026thinsp;=\u0026thinsp;102 high-risk HNSCC patients treated with postoperative RT) was downloaded from the Gene Expression Omnibus (GEO).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eConstruction of the LRSI\u003c/h3\u003e\n\u003cp\u003eWe assembled 547 candidate LRGs from eight Molecular Signatures Database (MSigDB) gene sets related to glycolysis, lactate metabolism, and lactic acidosis (HALLMARK_GLYCOLYSIS; GOMF_LACTATE_TRANSMEMBRANE_TRANSPORTER_ACTIVITY; HP_LACTIC_ACIDOSIS; HP_INCREASED_CIRCULATING_LACTATE_CONCENTRATION; HP_INCREASED_CIRCULATING_LACTATE_DEHYDROGENASE_CONCENTRATION; HP_ABNORMAL_CIRCULATING_LACTATE_DEHYDROGENASE_CONCENTRATION; KEGG_MEDICUS_REFERENCE_GLYCOLYSIS; WP_AEROBIC_GLYCOLYSIS), supplemented by a manually curated list of established lactylation writers, erasers, and readers [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Univariable Cox regression for OS was performed separately in the RT and non-RT cohorts; genes with false discovery rate (FDR)\u0026thinsp;\u0026lt;\u0026thinsp;0.05 in the RT cohort and FDR\u0026thinsp;\u0026gt;\u0026thinsp;0.20 in the non-RT cohort were retained as RT-specific candidates (n\u0026thinsp;=\u0026thinsp;11). These were further filtered by DDR pathway correlation (|Spearman ρ| \u0026gt; 0.30), yielding four genes.\u003c/p\u003e \u003cp\u003eGene expression values were z-score standardized within the RT cohort before model fitting. LASSO-penalized Cox regression with ten-fold cross-validation (CV) was performed using the \u0026ldquo;glmnet\u0026rdquo; R package; λ\u003csub\u003e1se\u003c/sub\u003e was selected to yield a parsimonious model. The resulting three-gene signature was:\u003c/p\u003e \u003cp\u003eLRSI\u0026thinsp;=\u0026thinsp;0.241 \u0026times; \u003cem\u003eCARS2\u003c/em\u003e\u0026thinsp;+\u0026thinsp;0.239 \u0026times; \u003cem\u003eKARS1\u003c/em\u003e\u0026thinsp;\u0026minus;\u0026thinsp;0.175 \u0026times; \u003cem\u003eBCL6\u003c/em\u003e\u003c/p\u003e \u003cp\u003ewhere expression values denote z-scored logCPM. Patients were classified as RS or RR using the RT-cohort median LRSI (\u0026minus;\u0026thinsp;0.021) as the cutoff. The same coefficients and cutoff were applied to the non-RT cohort.\u003c/p\u003e\n\u003ch3\u003ePredictive value assessment and robustness analyses\u003c/h3\u003e\n\u003cp\u003eKM analysis with log-rank tests compared OS between RS and RR within each treatment arm, and between RT and non-RT within each LRSI stratum. Cox models with a multiplicative RT \u0026times; LRSI interaction term tested for treatment-effect modification. To address confounding by treatment indication, PSM and IPTW-ATT were performed within each LRSI stratum using the \u0026ldquo;MatchIt\u0026rdquo; and \u0026ldquo;WeightIt\u0026rdquo; R packages, adjusting for age, sex, stage, grade, smoking history, and human papillomavirus (HPV) status; covariate balance was verified by standardized mean differences (SMD). RCS Cox models evaluated the continuous relationship between LRSI and the RT effect. Further sensitivity analyses included multivariable Cox regression adjusting for HPV status alone, for surgery type, and for both surgery type and chemotherapy receipt.\u003c/p\u003e\n\u003ch3\u003eExternal cohort analysis\u003c/h3\u003e\n\u003cp\u003eIn the GSE67614 cohort, the TCGA-derived LRSI coefficients were retained, while gene expression values were z-score normalized within the cohort to account for platform differences (microarray vs. RNA-Seq). Patients were dichotomized at the cohort-specific LRSI median. LRSI was assessed as both a continuous (per 1-SD) and dichotomized (RS/RR) predictor of locoregional recurrence. Discriminative ability was quantified by receiver operating characteristic (ROC) analysis using the \u0026ldquo;pROC\u0026rdquo; R package. Logistic regression models, with and without adjustment for T stage, N stage, and margin status, estimated the association between LRSI and recurrence risk. LRSI distributions were compared across failure phenotypes (no evidence of disease [NED], locoregional recurrence, and distant metastasis) using Kruskal-Wallis tests.\u003c/p\u003e\n\u003ch3\u003eImmune infiltration, SASP, and drug sensitivity analyses\u003c/h3\u003e\n\u003cp\u003eImmune cell abundance in the RT cohort was estimated using MCPcounter. A lactylation-informed DDR bridge score integrating glycolysis, hypoxia, and DNA repair activities was computed by single-sample gene set enrichment analysis (ssGSEA) using the \u0026ldquo;GSVA\u0026rdquo; R package. SASP activity was quantified via ssGSEA using senescence- and SASP-related gene sets (SenMayo, Reactome SASP, and Reactome cellular senescence). Predicted drug sensitivity was estimated with the \u0026ldquo;oncoPredict\u0026rdquo; R package using the Genomics of Drug Sensitivity in Cancer 2 (GDSC2) training panel; differential sensitivity between RS and RR was assessed by Wilcoxon tests with FDR correction.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eContinuous variables were compared using Wilcoxon rank-sum or Kruskal-Wallis tests; categorical variables using chi-square or Fisher exact tests. Survival was analyzed with KM curves, log-rank tests, and Cox proportional hazards regression. All \u003cem\u003eP\u003c/em\u003e values were two-sided (α\u0026thinsp;=\u0026thinsp;0.05). Analyses were performed in R version 4.5.1.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003eIdentification of RT-specific lactylation-related genes and construction of the LRSI\u003c/h2\u003e\n \u003cp\u003eOf 520 TCGA-HNSCC patients, 511 met the inclusion criteria (RT, n\u0026thinsp;=\u0026thinsp;308; non-RT, n\u0026thinsp;=\u0026thinsp;203; 215 OS events; Fig. \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA). Baseline characteristics are summarized in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. From 547 candidate LRGs, 11 were identified as RT-specific prognostic genes (FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05 in the RT cohort, FDR\u0026thinsp;\u0026gt;\u0026thinsp;0.20 in the non-RT cohort), of which four passed DDR pathway correlation filtering (|Spearman \u0026rho;| \u0026gt; 0.30; Fig. \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB). LASSO-penalized Cox regression with \u0026lambda;\u003csub\u003e1se\u003c/sub\u003e yielded a three-gene signature (Fig. \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC\u0026ndash;D):\u003c/p\u003e\n \u003cp\u003eLRSI\u0026thinsp;=\u0026thinsp;0.241 \u0026times; \u003cem\u003eCARS2\u003c/em\u003e\u0026thinsp;+\u0026thinsp;0.239 \u0026times; \u003cem\u003eKARS1\u003c/em\u003e\u0026thinsp;\u0026minus;\u0026thinsp;0.175 \u0026times; \u003cem\u003eBCL6\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eCARS2\u003c/em\u003e and \u003cem\u003eKARS1\u003c/em\u003e carried positive coefficients (higher expression associated with radioresistance), while \u003cem\u003eBCL6\u003c/em\u003e carried a negative coefficient (higher expression associated with radiosensitivity). Patients were dichotomized at the RT-cohort median LRSI (\u0026minus;\u0026thinsp;0.021) into RS (n\u0026thinsp;=\u0026thinsp;154) and RR (n\u0026thinsp;=\u0026thinsp;154) groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eE).\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eBaseline clinicopathological characteristics of the TCGA-HNSC derivation cohort according to radiotherapy status\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"6\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eCharacteristic\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eOverall (N\u0026thinsp;=\u0026thinsp;511)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eNo RT (N\u0026thinsp;=\u0026thinsp;203)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eRT (N\u0026thinsp;=\u0026thinsp;308)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c6\"\u003e\n \u003cp\u003eSMD\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e511\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e203\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e308\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eAge at diagnosis, mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e61.26 (11.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e63.56 (12.51)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e59.74 (11.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.323\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eMale sex, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e378 (74.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e135 (66.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e243 (78.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.281\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eHistologic grade, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.443\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eG1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e61 (11.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e36 (17.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e25 (8.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eG2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e297 (58.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e120 (59.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e177 (57.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eG3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e124 (24.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e44 (21.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e80 (26.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eG4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e7 (1.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e7 (2.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eGX\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e18 (3.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e3 (1.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e15 (4.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eUnknown\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e4 (0.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e4 (1.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eAJCC stage, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.763\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e27 (5.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e17 (8.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e10 (3.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e70 (13.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e51 (25.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e19 (6.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eIII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e81 (15.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e41 (20.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e40 (13.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eIV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e259 (50.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e81 (39.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e178 (57.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eUnknown\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e74 (14.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e13 (6.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e61 (19.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePathologic T stage, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.506\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eT1-2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e180 (35.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e95 (46.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e85 (27.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eT3-4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e269 (52.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e98 (48.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e171 (55.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eUnknown\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e62 (12.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e10 (4.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e52 (16.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePathologic N stage, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.384\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eN0-1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e240 (47.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e118 (58.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e122 (39.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eN2-3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e172 (33.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e51 (25.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e121 (39.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eUnknown\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e99 (19.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e34 (16.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e65 (21.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eSmoking history, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e0.039\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.216\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eNever\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e114 (22.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e43 (21.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e71 (23.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eEver\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e385 (75.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e151 (74.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e234 (76.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eUnknown\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e12 (2.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e9 (4.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e3 (1.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eHPV status, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\n \u003cp\u003e0.051\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\n \u003cp\u003e0.227\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e409 (80.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e170 (83.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e239 (77.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e71 (13.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e19 (9.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e52 (16.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eUnknown\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e31 (6.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e14 (6.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e17 (5.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cp\u003eValues are presented as mean (SD) or n (%). P values and SMDs were recalculated for the compact main-text version. T stage was collapsed as T1-2, T3-4, and Unknown; N stage as N0-1, N2-3, and Unknown; smoking history as Never, Ever, and Unknown.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eLRSI predicts differential RT benefit\u003c/h2\u003e\n \u003cp\u003eWithin the RT cohort, RR patients had significantly worse OS than RS patients (Fig. 2A; HR\u0026thinsp;=\u0026thinsp;2.04, RR vs. RS, 95% CI 1.40\u0026ndash;2.97, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). No such difference was observed in the non-RT cohort (Fig. 2B; HR\u0026thinsp;=\u0026thinsp;0.85, 95% CI 0.56\u0026ndash;1.27, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.41), indicating that the prognostic value of the LRSI was confined to RT-treated patients. When stratified by LRSI group, RS patients derived substantial OS benefit from RT (Fig. 2C; HR\u0026thinsp;=\u0026thinsp;0.45, 95% CI 0.30\u0026ndash;0.67, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), whereas RR patients did not (Fig. 2D; HR\u0026thinsp;=\u0026thinsp;0.98, 95% CI 0.68\u0026ndash;1.42, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.93). A significant multiplicative interaction between RT and LRSI group confirmed the predictive nature of this association (\u003cem\u003eP\u003c/em\u003e\u003csub\u003einteraction\u003c/sub\u003e = 0.0035).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003eRobustness of the predictive effect\u003c/h2\u003e\n \u003cp\u003eAfter IPTW-ATT weighting in the RS subgroup, all covariate SMDs fell below 0.10 (Fig. 3A) and the RT benefit remained significant (Fig. 3B; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.002). The RS subgroup showed consistent RT benefit across all analytical approaches: unadjusted HR\u0026thinsp;=\u0026thinsp;0.45 (95% CI 0.30\u0026ndash;0.67); propensity-based methods incorporating age, sex, stage, grade, smoking, and HPV status yielded PSM HR\u0026thinsp;=\u0026thinsp;0.49 (95% CI 0.32\u0026ndash;0.77) and IPTW-ATT HR\u0026thinsp;=\u0026thinsp;0.46 (95% CI 0.28\u0026ndash;0.74); multivariable Cox regression adjusting for HPV status gave HR\u0026thinsp;=\u0026thinsp;0.40 (95% CI 0.26\u0026ndash;0.62); adjustment for surgery type yielded HR\u0026thinsp;=\u0026thinsp;0.35 (95% CI 0.21\u0026ndash;0.58); and a fully adjusted model including both surgery and chemotherapy receipt yielded HR\u0026thinsp;=\u0026thinsp;0.30 (95% CI 0.18\u0026ndash;0.52). In the RR subgroup, no significant RT benefit was detected under any approach (unadjusted HR\u0026thinsp;=\u0026thinsp;0.98, IPTW-ATT HR\u0026thinsp;=\u0026thinsp;0.96, PSM HR\u0026thinsp;=\u0026thinsp;0.84; all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.20; Fig. 3C). RCS analysis confirmed a continuous decrease in RT benefit with increasing LRSI, with HR\u0026thinsp;\u0026lt;\u0026thinsp;1 confined to the lower end of the distribution (spline interaction LRT \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.016; Fig. 3D).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003eAssociation of LRSI with locoregional failure in an independent post-RT cohort\u003c/h2\u003e\n \u003cp\u003eTo evaluate generalizability, we applied the TCGA-derived LRSI coefficients to the GSE67614 cohort (n\u0026thinsp;=\u0026thinsp;102 high-risk HNSCC patients treated with postoperative RT). The LRSI discriminated locoregional recurrence with an AUC of 0.73 (95% CI 0.62\u0026ndash;0.84; Fig. \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). Patients who developed locoregional recurrence had significantly higher LRSI scores (Fig. \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003eB; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.001). Each 1-SD increase in LRSI was associated with a 2.10-fold higher recurrence risk (OR\u0026thinsp;=\u0026thinsp;2.10, 95% CI 1.23\u0026ndash;3.58, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.006), and this association persisted after adjustment for T stage, N stage, and margin status (adjusted OR\u0026thinsp;=\u0026thinsp;2.05, 95% CI 1.16\u0026ndash;3.61, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.013; Fig. \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003eC). Dichotomized LRSI showed a 4.21-fold increased recurrence risk in RR versus RS patients (OR\u0026thinsp;=\u0026thinsp;4.21, 95% CI 1.41\u0026ndash;12.59, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.010). The association was most pronounced for locoregional recurrence across failure phenotypes (Kruskal-Wallis \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.006; Fig. \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003eD), although the small number of distant metastasis events limits conclusions regarding endpoint specificity.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003eBiological characterization of LRSI-defined subgroups\u003c/h2\u003e\n \u003cp\u003eRR tumors exhibited significantly higher DDR bridge scores, reflecting elevated glycolysis\u0026ndash;hypoxia\u0026ndash;DNA repair activity (Fig. \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003eA; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). MCPcounter-based deconvolution revealed that RS tumors harbored greater infiltration of myeloid dendritic cells (DCs), B-lineage cells, T cells, and endothelial cells (all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01; Fig. \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003eB). SASP activity was elevated in RR tumors in both the RT and non-RT cohorts (both \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001; Fig. \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003eC). Drug sensitivity prediction identified subgroup-specific vulnerabilities: RS tumors showed greater predicted sensitivity to dasatinib, luminespib, and trametinib, whereas RR tumors showed greater sensitivity to docetaxel, obatoclax mesylate, and ulixertinib (FDR-corrected; Fig. \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003eD).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe LRSI reported here is a three-gene expression signature that identifies HNSCC patients who derive differential benefit from RT. Constructed through RT-specific gene selection, DDR pathway coupling, and LASSO regularization, the model comprising \u003cem\u003eCARS2\u003c/em\u003e, \u003cem\u003eKARS1\u003c/em\u003e, and \u003cem\u003eBCL6\u003c/em\u003e distinguished an RS subgroup with significant RT benefit from an RR subgroup without benefit. The significant treatment\u0026ndash;biomarker interaction and robustness across multiple confounding-adjustment methods\u0026mdash;including PSM, IPTW, HPV-adjusted, and surgery/chemotherapy-adjusted models\u0026mdash;support the LRSI as a candidate predictive biomarker in this disease. Separately, elevated LRSI was associated with locoregional failure in an independent post-RT cohort, providing prognostic corroboration in a distinct clinical setting.\u003c/p\u003e \u003cp\u003eDistinguishing predictive from prognostic biomarkers remains a central challenge [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Many radiosensitivity gene signatures were developed in patients who all received RT, without a non-RT control group, and are therefore prognostic by design [\u003cspan additionalcitationids=\"CR7\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. The LRSI was built to address this: genes were retained only if they carried prognostic significance in the RT cohort (FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05) but not in the non-RT cohort (FDR\u0026thinsp;\u0026gt;\u0026thinsp;0.20), thereby excluding genes with treatment-independent effects. The null association in non-RT patients and the significant interaction term reinforce the claim of predictive, rather than merely prognostic, utility.\u003c/p\u003e \u003cp\u003eExisting radiosensitivity models for HNSCC, including the RSI and multi-gene signatures from radiation-response gene sets, were generally trained on pan-cancer cell line data or lacked treatment-stratified evaluation, limiting their use as predictive tools[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The LRSI differs in being derived from a biologically defined candidate set\u0026mdash;lactylation-related genes\u0026mdash;rather than an agnostic transcriptome-wide screen, and in being evaluated through causal inference methods (PSM, IPTW) and continuous dose\u0026ndash;response modeling (RCS). The spline analysis demonstrated a graded decrease in RT benefit with rising LRSI, arguing against a purely artifactual dichotomous effect and strengthening the case for clinical relevance.\u003c/p\u003e \u003cp\u003e \u003cem\u003eCARS2\u003c/em\u003e encodes mitochondrial cysteinyl-tRNA synthetase, which produces cysteine persulfides essential for electron transport chain function [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Aminoacyl-tRNA synthetases have recently been recognized as non-canonical regulators of lactylation\u0026mdash;AARS1, for instance, acts as a lactate sensor and lactyltransferase\u0026mdash;and the positive LRSI coefficient of \u003cem\u003eCARS2\u003c/em\u003e is consistent with the idea that enhanced mitochondrial metabolic capacity contributes to radioresistance[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. \u003cem\u003eKARS1\u003c/em\u003e (lysyl-tRNA synthetase) has been linked to colorectal cancer diagnosis and promotes metastasis through the KARS1/ERK/paxillin axis while polarizing macrophages toward an M2 phenotype, an immunosuppressive property that aligns with the lower immune infiltration we observed in RR tumors[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. \u003cem\u003eBCL6\u003c/em\u003e is a transcriptional repressor of ATR and p53 that enables genotoxic stress evasion in solid tumors [\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Its negative coefficient\u0026mdash;meaning higher expression is associated with radiosensitivity\u0026mdash;may reflect DDR attenuation. Whether this link additionally involves B-cell-mediated immunity remains to be resolved at single-cell resolution.\u003c/p\u003e \u003cp\u003eThe lactylation\u0026ndash;DDR connection offers a biologically coherent framework for the LRSI. Histone Kla, driven by aerobic glycolysis, reprograms gene expression toward DNA repair and immune evasion, and lactylation of NBS1 has been shown to directly promote homologous recombination and chemoresistance [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Consistent with this, RR tumors in our cohort had significantly higher DDR bridge scores integrating glycolysis, hypoxia, and DNA repair pathway activity, supporting a model in which a lactylation-fueled metabolic-repair program underlies the radioresistant phenotype.\u003c/p\u003e \u003cp\u003eThe tumor immune microenvironment increasingly appears to modulate radiation response [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. RS tumors harbored greater infiltration of myeloid DCs, B-lineage cells, T cells, and endothelial cells, suggesting an immunologically active milieu that may potentiate the in situ vaccination effects of RT [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Conversely, RR tumors showed elevated SASP activity\u0026mdash;a phenotype linked to NF-κB-dependent cytokine-mediated radioresistance in HNSCC [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. The convergence of immune exclusion and SASP elevation in RR tumors points to a pro-tumorigenic microenvironment that blunts radiation efficacy.\u003c/p\u003e \u003cp\u003eDrug sensitivity prediction revealed subgroup-specific therapeutic vulnerabilities: RS tumors showed greater predicted sensitivity to dasatinib, luminespib (HSP90 inhibitor), and trametinib (MEK inhibitor), whereas RR tumors showed greater sensitivity to docetaxel and ulixertinib (ERK inhibitor). Although hypothesis-generating, these patterns raise the possibility that the LRSI could inform not only RT selection but also the choice of concurrent or adjuvant systemic therapy.\u003c/p\u003e \u003cp\u003eIn the GSE67614 cohort, applying the TCGA-derived LRSI coefficients with cohort-internal z-score standardization yielded acceptable discrimination for locoregional recurrence and maintained significance after adjustment for clinicopathological factors. The association was strongest for locoregional failure, consistent with the expectation that the signature captures local radiation response, though the small number of distant metastasis events limits conclusions on endpoint specificity. Because GSE67614 comprises uniformly RT-treated patients, it cannot directly test treatment\u0026ndash;biomarker interaction; rather, these data demonstrate that elevated LRSI is associated with locoregional failure in an independent post-RT population.\u003c/p\u003e \u003cp\u003eSeveral limitations should be acknowledged. First, this study was based on retrospective cohorts with non-randomized treatment assignment; despite PSM and IPTW, residual unmeasured confounding cannot be excluded, and predictive biomarkers ideally require validation in prospective randomized trials [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Second, the LRSI was derived from bulk RNA-Seq, which does not capture intratumoral heterogeneity or spatial context; single-cell and spatial transcriptomic studies may offer finer resolution. Third, the external cohort used a different endpoint (locoregional recurrence) and platform (microarray), and further validation in independent cohorts with time-to-event endpoints is needed. Fourth, our interpretation of the lactylation\u0026ndash;DDR axis relies on pathway-level inference without direct measurement of protein lactylation; experimental validation is required. Finally, the TCGA cohort is predominantly of European descent, and generalizability to other populations warrants investigation.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThe LRSI (\u003cem\u003eCARS2\u003c/em\u003e, \u003cem\u003eKARS1\u003c/em\u003e, \u003cem\u003eBCL6\u003c/em\u003e) is a three-gene signature that identifies HNSCC patients with differential RT benefit (RS: HR\u0026thinsp;=\u0026thinsp;0.45; RR: HR\u0026thinsp;=\u0026thinsp;0.98; \u003cem\u003eP\u003c/em\u003e\u003csub\u003einteraction\u003c/sub\u003e = 0.0035). This predictive effect was robust across multiple confounding-adjustment approaches. In an independent post-RT cohort, elevated LRSI was associated with locoregional failure (AUC\u0026thinsp;=\u0026thinsp;0.73), though this uniformly treated cohort could not directly assess treatment\u0026ndash;biomarker interaction. The signature captures a convergent axis linking glycolysis\u0026ndash;lactylation\u0026ndash;DDR activity, immune microenvironment composition, and SASP to radiation response. Prospective validation in randomized trials is warranted to establish clinical utility and guide personalized RT strategies in HNSCC.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch1\u003eEthics approval and consent to participate\u003c/h1\u003e\n\u003cp\u003eAll data were obtained from publicly available databases (TCGA and GEO). No additional ethical approval was required.\u003c/p\u003e\n\u003ch1\u003eAvailability of data and materials\u003c/h1\u003e\n\u003cp\u003eTCGA-HNSCC data are available from the GDC Data Portal (https://portal.gdc.cancer.gov/). The GSE67614 dataset is available from GEO (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE67614).\u0026nbsp;\u003c/p\u003e\n\u003ch1\u003eCompeting interests\u003c/h1\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch1\u003eFunding\u003c/h1\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003ch1\u003eAuthors\u0026rsquo; contributions\u003c/h1\u003e\n\u003cp\u003eXiong Yan: Conceptualization, Methodology, Formal analysis, Writing \u0026ndash; original draft, Writing \u0026ndash; review \u0026amp; editing, Supervision. Wang Zhe: Data curation, Software, Validation, Visualization, Writing \u0026ndash; review \u0026amp; editing.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEqual contribution: Xiong Yan and Wang Zhe contributed equally to this work and share first authorship.\u003c/p\u003e\n\u003ch1\u003eAcknowledgements\u003c/h1\u003e\n\u003cp\u003eWe acknowledge TCGA, and GEO for providing access to the datasets used in this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSung H, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209\u0026ndash;49.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBernier J, et al. Postoperative irradiation with or without concomitant chemotherapy for locally advanced head and neck cancer. N Engl J Med. 2004;350(19):1945\u0026ndash;52.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCooper JS, et al. Postoperative concurrent radiotherapy and chemotherapy for high-risk squamous-cell carcinoma of the head and neck. N Engl J Med. 2004;350(19):1937\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnson DE, et al. Head and neck squamous cell carcinoma. Nat Rev Dis Primers. 2020;6(1):92.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcLaughlin M, et al. Inflammatory microenvironment remodelling by tumour cells after radiotherapy. Nat Rev Cancer. 2020;20(4):203\u0026ndash;17.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTorres-Roca JF, et al. Prediction of radiation sensitivity using a gene expression classifier. Cancer Res. 2005;65(16):7169\u0026ndash;76.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEschrich SA, et al. A gene expression model of intrinsic tumor radiosensitivity: prediction of response and prognosis after chemoradiation. Int J Radiat Oncol Biol Phys. 2009;75(2):489\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBallman KV. Biomarker: Predictive or Prognostic? J Clin Oncol. 2015;33(33):3968\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCui Y, et al. Integrating Radiosensitivity and Immune Gene Signatures for Predicting Benefit of Radiotherapy in Breast Cancer. Clin Cancer Res. 2018;24(19):4754\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang D, et al. Metabolic regulation of gene expression by histone lactylation. Nature. 2019;574(7779):575\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen J, et al. Lactate and lactylation in cancer. Signal Transduct Target Ther. 2025;10(1):38.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu C, et al. Lactylation in radiosensitivity regulation: Mechanistic insights and recent advances. Semin Oncol. 2025;52(5):152394.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen H, et al. NBS1 lactylation is required for efficient DNA repair and chemotherapy resistance. Nature. 2024;631(8021):663\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang J, et al. Ubiquitous protein lactylation in health and diseases. Cell Mol Biol Lett. 2024;29(1):23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSimon RM, Paik S, Hayes DF. Use of archived specimens in evaluation of prognostic and predictive biomarkers. J Natl Cancer Inst. 2009;101(21):1446\u0026ndash;52.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEschrich SA, et al. Validation of a radiosensitivity molecular signature in breast cancer. Clin Cancer Res. 2012;18(18):5134\u0026ndash;43.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLu C, et al. Construction and validation of a prognostic nine-gene signature associated with radiosensitivity in head and neck squamous cell carcinoma. Clin Transl Radiat Oncol. 2023;43:100686.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYou P, et al. Radiation-sensitive genetic prognostic model identifies individuals at risk for radiation resistance in head and neck squamous cell carcinoma. J Cancer Res Clin Oncol. 2023;149(17):15623\u0026ndash;40.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkaike T, et al. Cysteinyl-tRNA synthetase governs cysteine polysulfidation and mitochondrial bioenergetics. Nat Commun. 2017;8(1):1177.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZong Z, et al. Alanyl-tRNA synthetase, AARS1, is a lactate sensor and lactyltransferase that lactylates p53 and contributes to tumorigenesis. Cell. 2024;187(10):2375\u0026ndash;e239233.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuh JH et al. Plasma Lysyl-tRNA Synthetase 1 (KARS1) as a Novel Diagnostic and Monitoring Biomarker for Colorectal Cancer. J Clin Med, 2020. 9(2).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNam SH, et al. Lysyl-tRNA synthetase-expressing colon spheroids induce M2 macrophage polarization to promote metastasis. J Clin Invest. 2018;128(11):5034\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRanuncolo SM, et al. Bcl-6 mediates the germinal center B cell phenotype and lymphomagenesis through transcriptional repression of the DNA-damage sensor ATR. Nat Immunol. 2007;8(7):705\u0026ndash;14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePhan RT, Dalla-Favera R. The BCL6 proto-oncogene suppresses p53 expression in germinal-centre B cells. Nature. 2004;432(7017):635\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Y et al. The oncoprotein BCL6 enables solid tumor cells to evade genotoxic stress. Elife, 2022. 11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNgwa W, et al. Using immunotherapy to boost the abscopal effect. Nat Rev Cancer. 2018;18(5):313\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchoetz U, et al. Early senescence and production of senescence-associated cytokines are major determinants of radioresistance in head-and-neck squamous cell carcinoma. Cell Death Dis. 2021;12(12):1162.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Head and neck squamous cell carcinoma, Lactylation, Radiosensitivity, Predictive biomarker, Tumor immune microenvironment","lastPublishedDoi":"10.21203/rs.3.rs-9250884/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9250884/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRadiotherapy (RT) is integral to the management of head and neck squamous cell carcinoma (HNSCC), yet no validated biomarker exists to predict which patients benefit most. Lactylation, a metabolic post-translational modification, has been linked to DNA damage repair (DDR) and therapeutic resistance, but no lactylation-based radiosensitivity biomarker has been established in HNSCC. Here, we analyzed RNA-Seq and clinical data from 511 TCGA-HNSCC patients (308 RT, 203 non-RT) and constructed a three-gene lactylation-related radiosensitivity index (LRSI) comprising CARS2, KARS1, and BCL6 via treatment-stratified gene selection and LASSO-Cox regression. Radiosensitive (RS) patients derived significant RT benefit (HR\u0026thinsp;=\u0026thinsp;0.45, 95% CI 0.30\u0026ndash;0.67, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), whereas radioresistant (RR) patients did not (HR\u0026thinsp;=\u0026thinsp;0.98, P\u0026thinsp;=\u0026thinsp;0.93; P_interaction\u0026thinsp;=\u0026thinsp;0.0035). These results were robust across propensity score matching, inverse probability of treatment weighting, HPV-adjusted, and surgery/chemotherapy-adjusted analyses. In the independent GSE67614 post-RT cohort (n\u0026thinsp;=\u0026thinsp;102), the LRSI discriminated locoregional recurrence with an AUC of 0.73 (adjusted OR\u0026thinsp;=\u0026thinsp;2.05, P\u0026thinsp;=\u0026thinsp;0.013). Biological characterization revealed that RR tumors exhibited elevated DDR activity, reduced immune infiltration, and heightened senescence-associated secretory phenotype. The LRSI is a predictive signature linking lactylation, DDR, and the tumor immune microenvironment to differential RT benefit in HNSCC. Prospective validation is warranted.\u003c/p\u003e","manuscriptTitle":"A Lactylation-Related Radiosensitivity Index Predicts Differential Radiotherapy Benefit in Head and Neck Squamous Cell Carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-21 01:31:15","doi":"10.21203/rs.3.rs-9250884/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-04-13T09:32:12+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-04-03T10:54:46+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-28T12:07:45+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-28T12:07:16+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Oncology","date":"2026-03-28T07:54:06+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6ebbae77-3c6a-4230-87e7-f95a4cfec3c8","owner":[],"postedDate":"April 21st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-21T01:31:15+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-21 01:31:15","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9250884","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9250884","identity":"rs-9250884","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}